A semantic equivalent text generation method based on familiar word constraint

Abstract

The application discloses a kind of semantic equivalent text generation methods based on familiar word constraint. Among them, the method includes: obtaining user familiar word set, and based on the familiar word set, construct structured limited expression space, for limiting the range of vocabulary and expression structure range that can be used in text generation process;Target text is carried out semantic structure analysis, and the target semantic structure representation is generated;In the structured limited expression space, semantic matching solution is executed, and text generation problem is converted into limited semantic solution problem, at least one expression path that satisfies the target semantic structure representation is constructed under the condition of the limited expression space;Expression candidate set is generated based on the expression path;The semantic consistency determination is carried out to the expression candidate set, and the semantic equivalent expression variant set that satisfies consistency condition is output.The application solves the technical problem that existing text generation technology is difficult to structure solution for target semantics within limited vocabulary range.

Description

Technical Field

[0001] This invention relates to the field of natural language processing and constrained text generation control technology, and more specifically, to a method for generating semantically equivalent text based on familiar word constraints. Background Technology

[0002] Existing text generation systems mostly rely on open-ended language models for probabilistic predictive generation, making it difficult to precisely control the vocabulary range during the generation process. In language learning or low cognitive load scenarios, open-ended generation suffers from the following problems: the output text may contain vocabulary beyond the user's grasp; it cannot guarantee that the generated expression is semantically consistent with the target text; it lacks a mechanism for generating systematic equivalent variants within a limited expression domain; and it cannot form a controlled set of expression variants.

[0003] In language learning or low cognitive load applications, existing generation methods struggle to stably construct a set of expression variants that are semantically consistent with the target text within a limited vocabulary.

[0004] Therefore, there is a need for a method that can solve the target semantics and generate a set of semantically equivalent expression variants within a structured and constrained expression space under the constraint of user-familiar words. Summary of the Invention

[0005] This invention provides a semantically equivalent text generation method based on familiar word constraints, which at least solves the technical problem that existing text generation technologies are unable to structurally solve the target semantics within a limited vocabulary.

[0006] According to one aspect of the present invention, a method for generating semantically equivalent text based on familiar word constraints is provided, comprising: obtaining a user's familiar word set, and constructing a structured restricted expression space based on the familiar word set to limit the vocabulary range and expression structure range that can be used in the text generation process; performing semantic structure parsing on the target text to generate a target semantic structure representation; performing semantic matching within the structured restricted expression space to transform the text generation problem into a restricted semantic solving problem, and constructing at least one expression path that satisfies the target semantic structure representation under the constraints of the restricted expression space; generating an expression candidate set based on the expression path; performing semantic consistency determination on the expression candidate set, and outputting a set of semantically equivalent expression variants that satisfy the consistency condition.

[0007] In this embodiment of the invention, by constructing a structured restricted expression space and executing a semantic matching solution mechanism within that space, a set of expression variants consistent with the semantics of the target text is generated under restricted conditions, thereby solving the technical problem that existing text generation technologies are unable to perform structured solutions to the target semantics within a limited vocabulary range. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0009] Figure 1 This is a flowchart of an optional semantically equivalent text generation method based on familiar word constraints according to an embodiment of the present invention;

[0010] Figure 2 This is an architecture diagram of an optional semantically equivalent text generation system based on familiar word constraints according to an embodiment of the present invention;

[0011] Figure 3 This is an optional structure construction mechanism for a restricted expression space (Ω) according to an embodiment of the present invention;

[0012] Figure 4 This is an optional structural semantic solution mechanism within a restricted expression space (Ω) according to an embodiment of the present invention;

[0013] Figure 5 According to an embodiment of the present invention, an optional process is to derive and expand multiple equivalent expression structure variants from a solved structural semantic core within a restricted expression space (Ω).

[0014] Figure 6 This is a flowchart of another semantically equivalent text generation method based on familiar word constraints according to an embodiment of the present invention;

[0015] Figure 7 This is a flowchart of another semantically equivalent text generation method based on familiar word constraints according to an embodiment of the present invention;

[0016] Figure 8 A schematic diagram of the structure of a computer device suitable for implementing embodiments of the present disclosure is shown. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] According to an embodiment of the present invention, a method embodiment for generating semantically equivalent text based on familiar word constraints is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0020] This application provides a method for generating semantically equivalent text variants within a constrained lexical space, applicable to language learning systems, personalized reading systems, text-controlled generation systems, and intelligent assisted expression systems. This method constructs a structured constrained expression space and executes a semantic matching mechanism within this space to generate a set of expression variants semantically consistent with the target text under constrained conditions.

[0021] Figure 1 This is a semantically equivalent text generation method based on familiar word constraints according to embodiments of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0022] Step S102: Construction of the structured restricted expression space.

[0023] Obtain the user's existing vocabulary set F, which is a set of usable vocabulary determined through recording or evaluation. Construct a structured restricted expression space Ω based on the vocabulary set F. The structured restricted expression space Ω is represented as a quadruple structure Ω = (V, E, C, T), where V represents the set of familiar word nodes, E represents the legal combination relationship edges between nodes, C represents the set of constraint rules, and T represents the set of syntactic structure templates. Node V includes part-of-speech tags, semantic tags, syntactic role tags, and combinable rule identifiers. The set of constraint rules C includes part-of-speech matching constraints, dependency relation constraints, semantic role consistency constraints, structural depth restriction rules, and familiar word proportion control rules. The restricted expression space Ω is a closed expression domain; in subsequent solution processes, no vocabulary generation models outside this space are called for probabilistic expansion.

[0024] Step S104: Target semantic structure abstraction and structural representation construction.

[0025] Semantic parsing is performed on the input target text to extract its semantic structure features, and a target semantic structure representation S is constructed. The semantic structure representation S is denoted as S=(A, R, M, L), where A represents the set of event subjects, R represents the behavior and relation structure, M represents the set of modifying attributes, and L represents the semantic hierarchy structure. The semantic structure representation includes event subject information, behavior relation mapping, time and space attributes, semantic dependency graph, and information hierarchy structure. The semantic structure representation S serves as the matching target for the subsequent constrained solution process.

[0026] Step S106: Solve the semantics within the constrained representation space.

[0027] A semantic matching solution process is performed within the restricted expression space Ω. This process is modeled as a constraint satisfaction problem. An expression path P is retrieved or constructed in Ω, consisting of a sequence of nodes and their combination relationships. The expression path must satisfy the following conditions: all nodes belong to the set of familiar words V, the node combinations satisfy the set of syntactic templates T, satisfy the set of constraint rules C, and maintain a consistent mapping with the target semantic structure representation S at the semantic structure level. When there are semantic elements that cannot be directly expressed, a semantic decomposition process is performed, including decomposing complex semantic units into expressible sub-semantic units, mapping the decomposed semantic units to a legal combination structure within Ω, and triggering a backtracking mechanism to reconstruct the expression path if the mapping fails. Through the above solution process, an expression candidate set E={E1, E2, …, En} is generated.

[0028] Step S108: Determine the consistency between the structured representation of the expression vector and its semantics.

[0029] For each candidate expression Ei, an expression vector representation Ei={Wi, Gi, Di} is constructed, where Wi is the word sequence, Gi is the set of syntactic structure labels, and Di is the semantic dependency graph. Semantic consistency is determined on the expression vectors, including semantic structure mapping consistency verification, dependency relationship consistency verification, semantic role matching degree calculation, and structural hierarchy matching degree calculation. Only expressions that meet a preset consistency threshold are retained; the consistency threshold is an adjustable parameter. The semantic consistency determination can be implemented based on semantic structure matching, semantic dependency relationship matching, or semantic vector similarity calculation.

[0030] Step S110: Structural deformation is generated.

[0031] Structural transformation operations are performed on expressions that pass the consistency determination; the structural transformations include, but are not limited to, active-passive structure conversion, word order rearrangement, information splitting, information merging, sentence replacement, and expression hierarchy reconstruction; the transformation operations are still completed within the restricted expression space Ω and do not exceed the limitation of the set of familiar words F; a set of semantically equivalent expression variants G is generated.

[0032] Step S112, output control and sorting mechanism.

[0033] The expression set G is sorted or graded based on the following indicators: usage ratio of familiar words, structural complexity, semantic reconstruction magnitude, expression difference, and cognitive load estimate. The system can output a single optimal equivalent expression, a set of multiple expression variants, or output in a graded manner according to complexity level.

[0034] Compared with existing open text generation technologies, this invention has the following technical advantages: it enables computationally comprehensible expression generation under limited lexical constraints; it ensures that the generated text is consistent with the target semantic structure; it forms a set of expression variants within a limited expression space; it provides a stable and controllable semantic equivalence generation mechanism; and it improves the applicability of text generation in low cognitive load scenarios.

[0035] In this embodiment, the semantic generation process is only allowed to be executed within the restricted expression space Ω, and any generation path outside the expression space is considered an infeasible path.

[0036] Figure 2 The overall structural architecture of the equivalent expression generation system according to the embodiments of this application includes an input processing module, a restricted expression space construction module, a structural semantic solving operator, and an equivalent expression variant generation module. Figure 3 The structural construction mechanism of the restricted expression space (Ω) is shown, including the familiar word set layer (F), the structural constraint rule layer (R), and the template and structure combination layer (T). Figure 4The structural semantic solution mechanism within a constrained expression space (Ω) is illustrated, in which constrained expression elements are processed by structural semantic solution operators to obtain a structural semantic core and generate equivalent expression variants. Figure 5 The process of deriving and expanding multiple equivalent structural variants of expression from a solved structural semantic core is shown within a constrained expression space (Ω).

[0037] Figure 6 This invention illustrates another semantically equivalent text generation method based on familiar word constraints, applied to the aforementioned equivalent expression generation system, such as... Figure 6 As shown, the method includes the following steps:

[0038] Step S602: Construct a restricted expression space based on the user's familiar word set, which includes word nodes, combination relationships, constraint rules and syntactic templates, in order to limit the expression range of subsequent text generation.

[0039] First, obtain the user's set of familiar words F. The set of familiar words F is a set of usable words determined by the user's historical learning records, vocabulary assessment results or ability assessment models, which is used to represent the range of vocabulary that the user can currently understand and express.

[0040] Based on the set of familiar words F, a set of lexical nodes V is constructed. Specifically, each familiar word is mapped to a corresponding lexical node Vi, and each lexical node is assigned a multi-dimensional attribute label. The attribute label includes at least: part-of-speech tag posi, semantic category tag semi, syntactic role tag syni, and combination rule identifier combi, so that the lexical node not only has a lexical identifier, but also has the ability to participate in the construction of syntactic structure.

[0041] Based on this, a set E of node combination relations is constructed. For example, based on the part-of-speech matching relationship between lexical nodes, a first type of connection relationship that satisfies basic grammatical rules is established; based on dependency parsing rules or preset dependency relation templates, a second type of connection relationship reflecting dependency relationships such as subject-predicate, verb-object, and modifier is established; based on the semantic compatibility between semantic category tags, a third type of semantically combinable connection relationship is established; the above multiple relationships are integrated to form a set E of legal combination relationship edges between nodes, thereby limiting the structural connection methods that can be formed between lexical nodes.

[0042] Furthermore, a set of constraint rules C is constructed. This set of constraint rules C is generated based on linguistic rules and expression control requirements, and includes at least: part-of-speech matching constraints based on part-of-speech tagging, used to limit the legal combinations between different parts of speech; dependency constraints based on dependency relations, used to ensure that the expression structure satisfies syntactic dependencies; semantic role consistency constraints based on semantic tags, used to ensure the matching relationship between semantic roles in the expression; structural depth restriction rules based on expression complexity control, used to constrain the hierarchical depth of the expression structure; and familiar word ratio control rules based on the user's familiarity with words, used to limit the scope of vocabulary usage in the expression. Through the combined effect of these rules, the generation process of the expression path is constrained and limited.

[0043] Simultaneously, a set of syntactic structure templates T is constructed. Specifically, based on preset common sentence structure patterns or high-frequency syntactic structures obtained statistically from the corpus, syntactic structures are abstracted to form a set of templates; each template is defined as a type of legal syntactic structure framework, used to specify the arrangement order of lexical nodes and their syntactic role configuration relationships; the syntactic templates may include, but are not limited to, subject-verb-object structure templates, subject-copula-predicate structure templates, coordinate structure templates, and nested clause structure templates.

[0044] Finally, based on the set of lexical nodes V, the set of node combination relationships E, the set of constraint rules C, and the set of syntactic structure templates T, a structured restricted expression space Ω=(V, E, C, T) is constructed. The expression space is a closed expression domain. In the subsequent text generation process, expression path construction is only allowed within this space, and lexical or generation models outside the space are not called for expansion.

[0045] Step S604: Perform multi-layer semantic parsing on the input text and construct a target semantic structure representation for matching.

[0046] First, the input target text is preprocessed, including word segmentation, part-of-speech tagging, and dependency parsing, to obtain the basic syntactic structure representation of the text. Word segmentation is used to divide the text into a sequence of lexical units, part-of-speech tagging is used to determine the part-of-speech category of each lexical unit, and dependency parsing is used to identify the syntactic dependencies between lexical units, thereby forming an initial syntactic dependency structure.

[0047] Based on the obtained basic syntactic structure, the event subject set A is extracted. Specifically, based on the dependency parsing results and semantic role labeling rules, entity words or phrases that assume the roles of event initiator, receiver, or participant are identified from the text, and classified and merged according to semantic roles to form the event subject set A, where each subject element is associated with at least its corresponding semantic role label and its position identifier in the syntactic structure.

[0048] Furthermore, a behavior and relation structure R is constructed. Specifically, this includes: identifying the core behavior in a sentence based on the verb headword, and determining the interaction between the behavior and the subject by combining dependency relations; at the same time, identifying the relational information expressed in prepositional structures, conjunction structures, or predicate complement structures, and unifying the connection between behavior and relation as a structured relation set R, thereby characterizing the behavioral logic and inter-entity relationships in the text.

[0049] Further, the set of modifying attributes M is extracted. Specifically, modifying components that limit or supplement the subject or behavior are identified from the text, including time attributes, spatial attributes, degree attributes, manner attributes, and state attributes; the modifying attributes are extracted through adverbial, attributive, or complement structures in dependency relations, and an association mapping is established with the corresponding subject or behavior, thereby forming the set of modifying attributes M.

[0050] Furthermore, a semantic hierarchical structure L is constructed. Specifically, based on the syntactic dependency tree structure and the importance of semantic information, the information units in the text are hierarchically divided, with core semantic units as primary level nodes and subordinate modifying information as secondary level nodes, forming a semantic structure with hierarchical nesting relationships; the semantic hierarchical structure is used to represent the primary and secondary relationships and combination priorities of information.

[0051] Based on the above, the event subject set A, behavior and relation structure R, modifier attribute set M, and semantic hierarchy structure L are uniformly organized to construct the target semantic structure representation S=(A, R, M, L).

[0052] Furthermore, the target semantic structure representation is converted into a semantic dependency graph structure Gs=(Ns, Es), where Ns represents a set of semantic nodes, the semantic nodes correspond to subject, behavior and attribute elements, and Es represents semantic relation edges, which are used to represent the semantic dependency relationship and hierarchical relationship between nodes.

[0053] Finally, semantic mapping constraints are constructed based on the semantic dependency graph Gs, so that the target semantic structure representation S can serve as the matching target in the subsequent semantic matching solution process within the restricted expression space Ω, and can be used to guide the construction and selection of expression paths.

[0054] Step S606: Within the structured and constrained expression space, the text generation problem is transformed into a constrained path optimization problem, and the optimal expression path is selected through a scoring function.

[0055] First, the semantic matching solution process is modeled as a restricted mapping problem from a semantic structure graph to an expression structure graph. Specifically, the semantic dependency graph Gs=(Ns, Es) constructed in step S604 is used as the source semantic graph, and the corresponding expression structure graph Gp=(Np, Ep) is constructed in the restricted expression space Ω, where Np represents the sequence of expression nodes and Ep represents the structural connection relationship between nodes.

[0056] Based on this, a structural semantic solver Ψ is introduced to perform structural decomposition and hierarchical processing on the semantic dependency graph Gs, resulting in a structurally stable kernel Sk and a semantic extension layer Se. The structurally stable kernel Sk includes the core predicate nodes in the semantic dependency graph and their directly dependent key semantic role nodes, which are used to represent the main semantic structure. The semantic extension layer Se includes modifying attribute nodes, additional semantic relation nodes, and hierarchical nested structure nodes, which are used to represent the extended semantic information.

[0057] Based on the structurally stable kernel Sk, the core mapping process is preferentially performed in the constrained expression space Ω. Specifically, a set of candidate nodes matching the semantic labels of each semantic node in Sk is retrieved from the lexical node set V, and a candidate connection subgraph is constructed based on the node combination relation set E. Under the constraints of the syntactic template set T, the candidate nodes are structurally assembled to generate an initial expression skeleton Gp_core that satisfies the basic syntactic structure.

[0058] After the initial expression skeleton Gp_core is constructed, an incremental structural expansion process for the semantic expansion layer Se is executed. Specifically, this includes: mapping semantic nodes in Se to the expression structure graph Gp layer by layer according to the hierarchical relationship order in the semantic dependency graph Gs; during the mapping process, embedding extended semantics into the current expression path through node insertion, sub-path concatenation, or local structure replacement operations; and simultaneously, using a set of constraint rules C to perform real-time constraint control on the expansion operations, ensuring that the expanded structure still satisfies syntactic validity and expression constraints.

[0059] Furthermore, a bidirectional semantic consistency alignment mechanism is introduced during the construction of the expression structure. This includes: a forward mapping constraint from the semantic graph Gs to the expression graph Gp, ensuring that the expression path covers the target semantic structure; and a reverse consistency check from the expression graph Gp to the semantic graph Gs, detecting whether there are redundant or deviating structural components in the expression structure. Through this bidirectional alignment mechanism, a structural-level consistent mapping between the expression path and the target semantics is achieved.

[0060] During path generation, a hierarchical constraint propagation and dynamic adjustment mechanism is used for adjustment. Specifically, various constraints in the constraint rule set C are represented as local constraint functions acting on nodes and edges, and propagated layer by layer during path expansion, so that local constraints have a synergistic effect on the global structure. At the same time, a constraint violation metric function is introduced to quantitatively evaluate the parts of the path that violate constraints, and the path expansion strategy is dynamically adjusted based on the evaluation results, thereby achieving a balance between constraint strictness and expression feasibility.

[0061] Furthermore, a backtrackable multi-branch search mechanism is employed during the path search process. Specifically, multiple candidate branch paths are retained during each layer of structural expansion, and each branch is scored based on semantic matching degree, structural legality, and constraint satisfaction. When the current path cannot complete the full semantic mapping, a backtracking operation is triggered, returning to the upper-level structural node to reselect a candidate branch, in order to avoid solution failure caused by local optima.

[0062] When a semantic node in the semantic dependency graph Gs that cannot be directly mapped within the current constrained expression space Ω is detected, semantic decomposition is performed. Specifically, this includes: decomposing the semantic node into multiple sub-semantic units using the structural semantic solver Ψ, and mapping each sub-semantic unit to a different position in the expression structure graph Gp; then, structurally reorganizing the sub-semantic units based on the node combination relation set E and the syntactic template set T, thereby achieving indirect expression of complex semantics; if mapping still cannot be completed after decomposition, the mapping method of the structural stability kernel Sk or the syntactic template selection strategy is readjusted through a backtracking mechanism.

[0063] After completing the above hierarchical mapping, structural expansion, constraint propagation and backtracking optimization, we obtain a set of expression paths P={P1, P2, …, P2} that satisfy semantic consistency and structural constraints, and construct an expression candidate set E={E1, E2, …, En} accordingly.

[0064] Through the above mechanism, a hierarchical solution framework based on structural semantic solution operators is formed within the restricted expression space Ω. The constrained expression elements are preferentially mapped under the guidance of the structurally stable kernel, and the overall expression is constructed through the gradual embedding of the semantic extension layer, thereby realizing the computable solution of the target semantic structure and the generation of semantically equivalent expression paths.

[0065] Step S608: Perform structured representation and multidimensional consistency determination on the candidate expression set, and filter the expression results that meet the semantic equivalence condition.

[0066] First, a structured representation transformation is performed on each candidate expression Ei in the candidate expression set E={E1, E2, …, En} generated in step S606. Specifically, each candidate expression is mapped to a corresponding expression structure graph Gpi=(Npi,Epi), where Npi represents the sequence of lexical nodes in the expression, and Epi represents the syntactic connection relationship between nodes. Based on this, an expression vector representation Vi=(Wi, Gi, Di, Ki) is constructed, where Wi represents the lexical sequence; Gi represents the set of syntactic structure labels; Di represents the semantic dependency graph; and Ki represents the mapping substructure corresponding to the structural stability kernel Sk, used to characterize the semantic core preservation in the expression.

[0067] After obtaining the representation of the expression structure, a semantic back-mapping mechanism is introduced to perform reverse semantic parsing on the expression structure. Specifically, the expression structure graph Gpi is remapped to the semantic representation Si'=(Ai', Ri', Mi', Li'), and the corresponding semantic dependency graph Gsi' is constructed, thereby realizing the reverse projection from the expression space to the semantic space.

[0068] Based on this, a two-way semantic consistency determination is performed. Specifically, this includes: comparing the target semantic structure S with the semantic structure Si' obtained by reverse mapping, and calculating the semantic subject matching degree, behavior relationship matching degree, modification attribute matching degree, and semantic level matching degree to assess whether the expression fully covers the target semantics; at the same time, performing structural alignment analysis between the expression structure graph Gpi and the original semantic dependency graph Gs to detect whether there are semantic missing, redundant expressions, or structural offset phenomena in the expression path.

[0069] Furthermore, an evaluation mechanism for the preservation of structural stability kernels is used. Specifically, the mapping consistency between the core substructure Ki in the expression vector and the structural stability kernel Sk extracted in step S606 is compared, and a core structure preservation index is calculated to evaluate the stability of the expression at the semantic backbone level. When the core structure is shifted or lost, the candidate expression is determined to not meet the equivalence condition.

[0070] Furthermore, a validity check based on constraint consistency is performed. Specifically, it checks whether the expression structure graph Gpi satisfies the constraint rule set C and syntactic template set T in the restricted expression space Ω, including part-of-speech matching validity, dependency relation validity, structural depth restriction, and familiar word ratio constraint; for expressions that violate the constraints, they are quantitatively evaluated by calculating the constraint violation index.

[0071] Based on the aforementioned multidimensional judgment, a consistency scoring function Cons(Ei) is constructed. This consistency scoring function integrates at least the following factors: semantic structure matching degree, dependency relationship consistency, structural stability kernel preservation degree, and constraint satisfaction degree. The scoring results are compared with a preset threshold, and only candidate expressions that meet the consistency conditions are retained. A structural difference index is calculated among the expressions that pass the consistency judgment to measure the degree of difference in syntactic structure and expression mode among different expression paths. While ensuring semantic consistency, expressions with greater structural differences are preferentially retained to avoid homogenization of output results. Finally, the set of expressions E' that pass the consistency screening is obtained and serves as the input for the subsequent structural deformation generation step.

[0072] Through the above mechanism, structural-level semantic verification of the generated results within the restricted expression space Ω is achieved. Semantic back-mapping, bidirectional consistency alignment, and structural stability kernel preservation evaluation ensure that the expression results are not only controlled at the lexical level, but also maintain strict consistency with the target semantics at the semantic structure level, thereby guaranteeing the accuracy and stability of semantically equivalent expression generation.

[0073] Step S610: Without changing the semantics, perform various transformations on the expression structure to generate expression variants.

[0074] First, for the expression set E' obtained in step S608, select the expression structure graph Gpᵢ and its structurally stable kernel substructure Kᵢ corresponding to each candidate expression Eᵢ. The structurally stable kernel Kᵢ is set as the invariant constraint core in the deformation process, which is used to limit the semantic backbone structure to remain unchanged in the deformation process.

[0075] Based on this, the set of structural transformation operators Θ={θ1, θ2, …, θm} is used to perform different types of structural transformation operations on the expression structure. The structural transformation operators include at least the following types: word order adjustment operators based on syntactic structure rearrangement, used to change the order of lexical nodes in the expression structure; active / passive voice conversion operators based on voice transformation, used to map between different syntactic expression forms; structural splitting and merging operators based on information organization methods, used to split a single expression into multiple sub-expressions or combine multiple expressions into a composite expression; template transformation operators based on sentence substitution, used to perform structural substitution between different syntactic templates; and semantic nesting adjustment operators based on hierarchical reconstruction, used to adjust the presentation of the semantic hierarchical structure L in the expression.

[0076] Furthermore, for each candidate expression Ei, while keeping the structurally stable kernel Ki unchanged, its corresponding variable structure region is identified, that is, the extended subgraph region in the expression structure graph Gpi other than the core substructure; within the variable structure region, an applicable structural deformation operator θk is selected to perform replacement, rearrangement or reconstruction operations on the local structure, thereby generating a new expression structure graph Gpi'.

[0077] During the deformation process, a structural feasibility constraint control mechanism is introduced. Specifically, each structural deformation operation must meet the following conditions: the deformed expression structure still conforms to the set of node combination relations E in the restricted expression space Ω; the deformed structure satisfies the structural pattern defined by the syntactic template set T; the lexical nodes used in the deformation process still belong to the set of familiar words V; and the deformation operation does not disrupt the mapping consistency between the structural stability kernel Ki and the original structural stability kernel Sk.

[0078] Furthermore, after the structural deformation is generated, a fast semantic consistency check is performed on the generated expression structure Gpi'. Specifically, by calling the semantic back-mapping mechanism in step S608, the semantic back-parsing of the deformed expression is performed, and the matching degree between it and the target semantic structure S is calculated; only expression variants that meet the semantic consistency threshold are retained, thereby ensuring that the deformation operation does not introduce semantic shift.

[0079] Furthermore, a diversity control mechanism driven by structural dissimilarity is used to control diversity. Specifically, a structural dissimilarity index is calculated between the original expression structure Gpi and the deformed structure Gpi' to measure the degree of change in expression mode; while satisfying semantic consistency, expression variants with greater dissimilarity are preferentially retained to improve the diversity of output results.

[0080] Furthermore, during the multi-round structural deformation process, a combined deformation strategy is used for combination. Specifically, multiple structural deformation operators θk are applied sequentially to the same expression to form a composite deformation path; and constraint verification and semantic consistency review are performed after each round of deformation, thereby gradually generating a multi-level, multi-structure expression variant set.

[0081] Finally, based on the above structural deformation operation and filtering mechanism, a set of semantically equivalent expression variants G={G1,G2, …, Gm} is generated, in which each expression variant is consistent with the target semantics at the semantic structure level, but differs in syntactic structure, expression form and information organization.

[0082] The above method realizes controlled expression generation within a restricted expression space Ω. The semantic backbone is locked by a structural stability kernel, and the peripheral structure is controlled by a structural deformation operator. At the same time, the semantic consistency verification mechanism is combined to ensure that the generated result has structural diversity while maintaining semantic equivalence. This is different from traditional expression variant generation methods based on simple rule replacement or random generation.

[0083] Step S612: Sort and output the expression results based on multidimensional evaluation indicators.

[0084] First, for the set of semantically equivalent expression variants G={G1, G2, …, Gm} generated in step S610, a multi-dimensional evaluation vector Qi is constructed for each expression variant Gi to characterize the comprehensive attributes of the expression in terms of lexical use, structural features and cognitive load.

[0085] Specifically, the evaluation vector Qi includes at least the following metrics: Rvocab (proportion of familiar words used), used to characterize the proportion of words in the expression that belong to the set of familiar words F; Cstruct (structural complexity), used to characterize the hierarchical depth, nesting degree, and number of nodes of the expression structure graph Gpi'; Ddiff (expression difference), used to characterize the degree of structural difference between the expression and the original expression and other expression variants; Scons (semantic consistency preservation), used to characterize the consistency level of the expression with respect to the target semantic structure S; and Lload (cognitive load estimate), used to estimate the cognitive cost required for a user to understand the expression based on structural complexity and word distribution.

[0086] After obtaining the evaluation vectors for each expression variant, an output scoring function, Score_out(Gi), is constructed to comprehensively evaluate the expression variants. This scoring function is a multi-index weighted function, and the weights of each index can be dynamically adjusted according to the application scenario or user needs, thereby achieving adaptation to different output targets.

[0087] Furthermore, a user-adaptive-driven output strategy is introduced. Specifically, the weight parameters of each evaluation indicator in the scoring function are dynamically adjusted according to the user's language proficiency level or target application scenario. When the user prefers low cognitive load, the weight of the proportion of familiar words and low-complexity expressions is increased; when the user prefers expression diversity or training needs, the weight of expression difference and structural complexity is increased, thereby achieving personalized adaptation of the output results.

[0088] Furthermore, in some embodiments, a hierarchical output control mechanism can be utilized during the sorting process. Specifically, this includes: sorting expression variants according to their scoring results and selecting the highest-scoring expression as the optimal output; simultaneously, dividing expression variants into multiple difficulty levels based on structural complexity or cognitive load indicators, and outputting a set of expressions corresponding to each level; in scenarios requiring diverse outputs, selecting representative expressions from different structural categories or difference ranges to form a structurally diverse output set. Additionally, stability screening can be performed on expression variants with similar scores to avoid frequent fluctuations in output results due to minor score differences; furthermore, maintaining consistency in output style across multiple generation tasks improves system predictability and user experience consistency.

[0089] Finally, based on the scoring results and output strategy, an output result R is generated. The output result may include: a single optimal semantically equivalent expression, a set of multiple expressions ordered by priority, or a set of expression variants graded by difficulty.

[0090] The above method achieves closed-loop control of the entire process from expression generation to result output. Through multi-dimensional evaluation modeling, dynamic weight adjustment and hierarchical output strategy, the generated result not only meets the requirements of semantic equivalence, but also can adaptively balance readability, complexity and diversity according to user needs, thus distinguishing it from traditional text generation methods that only provide a single generated result.

[0091] Figure 7 This is yet another semantically equivalent text generation method based on familiar word constraints according to embodiments of this application, such as... Figure 7 As shown, it includes the following steps:

[0092] Step S702: Obtain the user's familiar word set F, and construct a structured restricted expression space Ω based on the familiar word set F to limit the range of vocabulary and expression structure that can be used in the text generation process;

[0093] For example, a set of node combination relations E is constructed based on the part-of-speech matching relationship, dependency relationship and semantic association relationship between the word nodes; a set of constraint rules C is constructed based on the part-of-speech matching relationship, dependency relationship and semantic role consistency; a set of syntactic structure templates T is constructed based on the preset syntactic structure or corpus statistical results; and the structured restricted expression space Ω is constructed based on the set of word nodes V, the set of node combination relations E, the set of constraint rules C and the set of syntactic structure templates T.

[0094] Step 704: Perform semantic structure parsing on the target text, extract semantic structure features, and construct the target semantic structure representation S;

[0095] For example, the target text is segmented, part-of-speech tagging is performed, and dependency parsing is performed to obtain the basic syntactic structure; based on the basic syntactic structure, the event subject set A, the behavior and relation structure R, the modifying attribute set M, and the semantic hierarchy structure L are extracted; based on the event subject set A, the behavior and relation structure R, the modifying attribute set M, and the semantic hierarchy structure L, the target semantic structure representation S is constructed; the target semantic structure representation S is converted into a semantic dependency graph Gs for subsequent semantic matching solution.

[0096] Step S706: Within the structured restricted expression space Ω, perform semantic matching solution based on the target semantic structure representation S, transform the text generation problem into a restricted semantic solution problem, and construct at least one expression path P that satisfies the target semantic structure representation S within the structured restricted expression space Ω;

[0097] For example, the semantic dependency graph Gs is decomposed using a structural semantic solver to obtain a structurally stable kernel Sk and a semantic extension layer Se; an initial expression skeleton Gp_core is constructed in the structured restricted expression space Ω based on the structurally stable kernel Sk; the initial expression skeleton Gp_core is structurally extended based on the semantic extension layer Se to obtain an expression path P; and an expression candidate set E is generated based on the expression path P.

[0098] In some embodiments, the semantic dependency graph Gs is decomposed using the structural semantic solver to obtain a structurally stable kernel Sk, including: extracting predicate nodes and their directly dependent semantic role nodes from the semantic dependency graph Gs to construct the structurally stable kernel Sk; matching candidate nodes in the lexical node set V based on the structurally stable kernel Sk, and constructing candidate connection structures by combining the node combination relation set E; assembling the candidate connection structures under the constraints of the syntactic structure template set T to obtain the initial expression skeleton Gp_core.

[0099] In some embodiments, the initial representation skeleton Gp_core is structurally expanded based on the semantic extension layer Se to obtain the representation path P, including: mapping the semantic nodes in the semantic extension layer Se to the initial representation skeleton Gp_core layer by layer; performing constraint propagation based on the constraint rule set C during the mapping process to dynamically adjust the representation structure; generating multiple candidate paths during the structural expansion process and filtering them based on constraint satisfaction and semantic matching degree; performing a backtracking operation when the current path does not meet the semantic mapping requirements and reselecting the expansion path; and finally obtaining the representation path P that satisfies the target semantic structure representation S.

[0100] Step S708: Generate an expression candidate set E based on the expression path P, and perform semantic consistency determination on the expression candidate set E to obtain an expression set E' that passes the consistency determination;

[0101] For example, each candidate expression in the expression candidate set E is converted into an expression structure graph Gp and an expression vector representation; semantic back mapping is performed on the expression structure graph Gp to obtain the corresponding semantic structure representation S'; the semantic structure representation S' is compared with the target semantic structure representation S to calculate semantic consistency; the expression candidate set E is filtered based on the semantic consistency to obtain the expression set E'.

[0102] In some embodiments, the semantic structure representation S' is compared with the target semantic structure representation S to calculate semantic consistency, including: calculating the consistency of event subject set, behavioral relationship, modification attribute consistency, and semantic hierarchy consistency; extracting the core substructure K in the expression structure graph Gp, and matching the core substructure K with the structural stability kernel Sk to calculate the structural stability kernel preservation degree; and filtering the expression candidates based on the semantic consistency and structural stability kernel preservation degree to obtain the expression set E'.

[0103] Step S710: Perform structural deformation generation based on the expression set E' to obtain a set of semantically equivalent expression variants G;

[0104] For example, based on the expression structure graph Gp corresponding to each expression, a structurally stable kernel Sk and a variable structural region are determined; while keeping the structurally stable kernel Sk unchanged, a structural transformation is performed on the variable structural region by applying a set of structural deformation operators Θ to obtain the transformed expression structure; a semantic consistency verification is performed on the transformed expression structure; and the set of semantically equivalent expression variants G is obtained based on the verification results.

[0105] Step S712: Based on the set of semantically equivalent expression variants G, a comprehensive evaluation and output control are performed to obtain the semantically equivalent expression result of the target semantics.

[0106] For example, an evaluation vector is constructed for each expression in the set of semantically equivalent expression variants G, the evaluation vector including the proportion of familiar words used, structural complexity, expression difference, and semantic consistency index; a scoring function is constructed based on the evaluation vector to score and rank the set of semantically equivalent expression variants G; a target expression is selected from the set of semantically equivalent expression variants G based on the scoring results and a preset output strategy; the target expression is output as the semantically equivalent expression result of the target semantics.

[0107] Figure 8 A schematic diagram of a computer device suitable for implementing embodiments of the present disclosure is shown. It should be noted that... Figure 8 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0108] like Figure 8 As shown, the computer device includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage section 1008 into a random access memory (RAM) 1003. The RAM 1003 also stores various programs and data required for system operation. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0109] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 1010 as needed so that computer programs read from it can be installed into storage section 1008 as needed.

[0110] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating semantically equivalent text based on familiar word constraints, characterized in that, include: Obtain a set of familiar words from users, and construct a structured restricted expression space based on the set of familiar words to limit the range of vocabulary and expression structure that can be used in the text generation process; Perform semantic structure parsing on the target text to generate a target semantic structure representation; Semantic matching is performed within the structured restricted expression space, transforming the text generation problem into a restricted semantic solution problem. Under the constraints of the restricted expression space, at least one expression path that satisfies the target semantic structure representation is constructed. An expression candidate set is generated based on the expression path, a semantic consistency determination is performed on the expression candidate set, and a set of semantically equivalent expression variants that meet the consistency condition is output.

2. The method according to claim 1, characterized in that, The structured restricted expression space includes: a set of familiar words, a set of rules for combining familiar words, and a set of syntactic structure templates.

3. The method according to claim 1, characterized in that, The structured restricted expression space is constructed using the following elements: a set of lexical nodes, node combination relationships, expression constraint rules, and syntactic structure templates.

4. The method according to claim 1, characterized in that, The expression constraint rules include at least one of the following: part-of-speech matching constraints, dependency relation constraints, semantic role consistency constraints, and expression structure depth restriction rules.

5. The method according to claim 1, characterized in that, The semantic matching solution process includes searching for expression paths that satisfy the target semantic structure representation within the restricted expression space.

6. The method according to claim 1, characterized in that, When there are semantic elements in the target semantic structure that cannot be directly expressed in the restricted expression space, the semantic elements are semantically decomposed, and the decomposed semantic units are remapped to the legal expression paths in the restricted expression space.

7. The method according to claim 1, characterized in that, The candidate set of expressions is obtained by generating multiple structurally different but semantically consistent expression paths within the restricted expression space.

8. The method according to claim 1, characterized in that, The semantic consistency determination includes semantic structure mapping consistency verification or semantic dependency relationship consistency verification.

9. The method according to claim 1, characterized in that, Structural deformation operations are performed on expressions that pass the semantic consistency determination to generate multiple semantically equivalent expression variants.

10. The method according to claim 1, characterized in that, The set of semantically equivalent expression variants is sorted or graded and output according to expression complexity, proportion of familiar words, or structural difference.

Images